Urban Institute R Graphics Guide
R is a powerful, open-source programming language and environment. R excels at data management and munging, traditional statistical analysis, machine learning, and reproducible research, but it is probably best known for its graphics. This .html file made with R Markdown contains examples and instructions for popular and lesser-known plotting techniques in R. Furthermore, it includes instructions on how to leverage the Urban Institute’s ggplot2 theme which will create near-publication-ready plots. If you have any questions, please don’t hesitate to contact Aaron Williams (awilliams@urban.org) or Kyle Ueyama (kueyama@urban.org).
Using the Urban Institute ggplot2 Theme
Depending on your operating system, run one of the following scripts once per R session:
Windows
source('https://raw.githubusercontent.com/UrbanInstitute/urban_R_theme/master/urban_theme_windows.R')Mac
source('https://raw.githubusercontent.com/UrbanInstitute/urban_R_theme/master/urban_theme_mac.R')This is an R script that makes ggplot2 output align more closely with the Urban Institute’s Data Visualization style guide.
- This script does not produce publication ready graphics. Visual styles must still be edited using your project/paper’s normal editing workflow.
- Exporting charts as a pdf will allow them to be more easily edited
- You may need to tweak pdf export options to preserve fonts. For example, in RStudio on OSX, check this option in the export pdf window:
The theme has been tested against
ggplot2 version 2.2.1. It will not function properly with older (< 2.0.0) versions ofggplot2- If it is not already installed, please install the free Lato font from Google fonts.
- If you’re on Windows, you’ll first need to install Ghostscript. You may need to have IT enter an admin password for this installation. Then, in R, tell R where your ghostscript file is.
Edit the file path if yours is in a different place
Sys.setenv(R_GSCMD="C:/Program Files/gs/gs9.05/bin/gswin32c.exe")`
Run this script once:
install.packages(c("ggplot2", "ggrepel", "RColorBrewer", "extrafont"))
library(extrafont)
port()
loadfonts()Loading and importing fonts may take a few minutes.
After the initial installation, to use Lato just load the library in each R session:
library(extrafont)Grammar of Graphics and Conventions
Hadley Wickham’s ggplot2 is based on Leland Wilkinson’s The Grammar of Graphics and Wickham’s A Layered Grammar of Graphics. The layered grammar of graphics is a structured way of thinking about the components of a plot, which then lend themselves to the simple structure of ggplot2.
- Data are what are visualizaed in a plot and mappings are directions for how data are mapped in a plot in a way that can be perceived by humans.
- Geoms are representations of the actual data like points, lines, and bars.
- Stats are statistical transformations that represent summaries of the data like histograms.
- Scales map values in the data space to values in the aesthetic space. Scales draw legends and axes.
- Coordinate Systems describe how geoms are mapped to the plane of the graphic.
- Facets break the data into meaningful subsets like small multiples.
- Themes control the finer points of a plot such as fonts, font sizes, and background colors.
More information: ggplot2: Elegant Graphics for Data Analysis
Tips and Tricks
ggplot2expects data to be in data frames. It is preferable for the data frames to be “tidy” with each variable as a column, each obseravtion as a row, and each observational unit as a separate table.dplyrandtidyrcontain concise and effective tools for “tidying” data.R allows function arguments to be called explicitly by name and implicitly by position. The coding examples in this guide only contain named arguments for clarity.
Continuous legends should be switched to vertical using
theme(legend.direction = "vertical").Graphics will sometimes render differently on different operating systems. This is because anti-aliasing is activated in R on Mac and Linux but not activated in R on Windows. This blog post outlines several fixes for this problem.
Most features of plots can be adjusted by adding
theme()to the end of a ggplot call. For example, a plot with a continuous legend would look like this:
ggplot(diamonds, aes(carat, price)) +
stat_binhex(aes(colour = ..count..)) +
theme(legend.position = "right",
legend.direction = "vertical")Bar Plots
One Color
ggplot(data = mtcars, mapping = aes(factor(cyl))) +
geom_bar() +
scale_y_continuous(expand = c(0, 0), limits = c(0, 20)) +
labs(title = "Number of Cars By Number of Cylinders",
caption = "Urban Institute",
x = "Number of Cylinders",
y = "Count")
Three Colors
This is identical to the previous plot except colors and a legend are added with fill = factor(cyl). Turning x into a factor with factor(cyl) skips 5 and 7 on the x-axis. Adding fill = cyl without factor() would have created a continuous color scheme and legend.
ggplot(data = mtcars, mapping = aes(x = factor(cyl), fill = factor(cyl))) +
geom_bar() +
scale_y_continuous(expand = c(0, 0)) +
labs(title = "Number of Cars By Number of Cylinders",
caption = "Urban Institute",
x = "Number of Cylinders",
y = "Count"
) 
Stacked Bar Plot
An additional aesthetic can easily be added to bar plots by adding fill = categorical variable to the mapping. Here, diamond quality subsets each bar showing the count of diamonds with each level of clarity.
ggplot(data = diamonds, mapping = aes(x = clarity, fill = cut)) +
geom_bar() +
scale_y_continuous(expand = c(0, 0), limits = c(0, 15000), labels = scales::comma) +
labs(
title = "Diamond Clarity",
subtitle = "Something Informative About Diamonds",
caption = "The Source of Diamond Data",
x = "Clarity",
y = "Count"
)
Stacked Bar Plot With Position = Fill
position = "fill" in geom_bar() changes the y-axis from count to the proportion of each bar.
#5 colors (stacked)
ggplot(data = diamonds, mapping = aes(x = clarity, fill = cut)) +
geom_bar(position = "fill") +
scale_y_continuous(expand = c(0, 0), labels = scales::percent) +
labs(title = "Better Cut Diamonds have Better Clarity",
subtitle = "Share of Diamonds with Different Qualities by Clarity of Cut",
caption = "The Source of Diamond Data",
x = "Clarity",
y = "Count")
Dodged Bar Plot
Subsetted bar charts in ggplot2 are stacked by default. position = "dodge" in geom_bar() expands the bar chart so the bars appear next to each other.
#5 colors (dodged)
ggplot(data = diamonds, mapping = aes(clarity, fill = cut)) +
geom_bar(position = "dodge") +
scale_y_continuous(expand = c(0, 0), limits = c(0, 6000), labels = scales::comma) +
labs(title = "Diamond Clarity",
subtitle = "Something Informative About Diamonds",
caption = "The Source of Diamond Data",
x = "Clarity",
y = "Count")
Lollipop plot/Cleveland dot plot
Lollipop plots and Cleveland dot plots are minimalist alternatives to bar plots. The key to both plots is to order the data based on the continuous variable using arrange() and then turn the discrete variable into a factor with the ordered levels of the continuous variable using mutate(). This step “stores” the order of the data.
Lollipop plot
mtcars %>%
rownames_to_column("model") %>%
arrange(mpg) %>%
mutate(model = factor(model, levels = .$model)) %>%
ggplot(aes(mpg, model)) +
geom_segment(aes(x = 0, xend = mpg, y = model, yend = model)) +
geom_point() +
scale_x_continuous(expand = c(0, 0), limits = c(0, max(mtcars$mpg) * 1.1)) +
labs(title = "Miles Per Gallon of Popular Cars",
subtitle = "1974 Motor Trend US magazine",
x = NULL,
y = "Miles Per Gallon",
caption = "Urban Institute") +
theme(axis.text.y = element_text(size = 8))
Cleveland dot plot
mtcars %>%
rownames_to_column("model") %>%
arrange(mpg) %>%
mutate(model = factor(model, levels = .$model)) %>%
ggplot(aes(mpg, model)) +
geom_point() +
scale_x_continuous(expand = c(0, 0), limits = c(0, max(mtcars$mpg) * 1.1)) +
labs(title = "Miles Per Gallon of Popular Cars",
subtitle = "1974 Motor Trend US magazine",
x = NULL,
y = "Miles Per Gallon",
caption = "Urban Institute") +
theme(axis.text.y = element_text(size = 8))
Dumbell plot
Scatter Plots
One Color Scatter Plot
Scatter plots are useful for showing relationships between two or more variables.
ggplot(data = diamonds, mapping = aes(x = carat, y = price)) +
geom_point() +
scale_y_continuous(expand = c(0, 0), labels = scales::dollar) +
labs(title = "Diamond Prices Increase With Size",
subtitle = "Diamond Prices in Dollars and Sizes in Carats",
caption = "Urban Institute",
x = "Carat",
y = "Price"
)
High-Density Scatter Plot with Transparency
Large numbers of observations can sometimes make scatter plots tough to interpret because points overlap. Adding alpha = with a number between 0 and 1 adds transparency to points and clarity to plots. Now it’s easy to see that jewelry stores are probably rounding up but not rounding down carats!
ggplot(data = diamonds, mapping = aes(x = carat, y = price)) +
geom_point(alpha = 0.05) +
scale_y_continuous(expand = c(0, 0), labels = scales::dollar) +
labs(title = "Diamond Prices Increase With Size",
subtitle = "Diamond Prices in Dollars and Sizes in Carats",
caption = "Urban Institute",
x = "Carat",
y = "Price"
)
Hex Scatter Plot
Sometimes transparency isn’t enough to bring clarity to a scatter plot with many observations. As n increases into the hundreds of thousands and even millions, geom_hex can be one of the best ways to display relationships between two variables.
ggplot(data = diamonds, mapping = aes(x = carat, y = price)) +
geom_hex(mapping = aes(fill = ..count..)) +
scale_y_continuous(expand = c(0, 0), labels = scales::dollar) +
scale_fill_gradientn(labels = scales::comma) +
labs(title = "Title",
subtitle = "geom_hex adds clarity to overlapping points",
x = "Carat",
y = "Price") +
theme(legend.position = "right",
legend.direction = "vertical")
Scatter Plots With Random Noise
Sometimes scatter plots have many overlapping points but a reasonable number of observations. geom_jitter adds a small amount of random noise so points are less likely to overlap. width and height control the amount of noise that is added. In the following before-and-after, notice how many more points are visible after adding jitter.
Before
ggplot(data = mpg, mapping = aes(x = displ, y = cty)) +
geom_point() +
scale_y_continuous(expand = c(0, 0)) +
labs(title = "Displacement and City MPG",
subtitle = "Cars With Less Displacement Generally Get Better City MPG",
caption = "Urban Institute",
x = "Displacement",
y = "City MPG"
)
After
set.seed(2017)
ggplot(data = mpg, mapping = aes(x = displ, y = cty)) +
geom_jitter(width = 0.2, height = 0.2) +
scale_y_continuous(expand = c(0, 0)) +
labs(title = "Displacement and City MPG",
subtitle = "Cars With Less Displacement Generally Get Better City MPG",
caption = "Urban Institute",
x = "Displacement",
y = "City MPG"
)
Scatter Plot with Counts
Another option is to use geom_count() to add a size dimension to overlapping points.
ggplot(data = mpg, mapping = aes(x = displ, y = cty)) +
geom_count() +
scale_y_continuous() +
labs(title = "Displacement and City MPG",
subtitle = "Cars With Less Displacement Generally Get Better City MPG",
caption = "Urban Institute",
x = "Displacement",
y = "City MPG"
)
Scatter Plots with Fill
A third aesthetic can be added to scatter plots. Here, color signifies the number of cylinders in each car. Before ggplot() is called, Cylinders is created using library(dplyr) and the piping operator %>%.
mtcars %>%
mutate(Cylinders = factor(cyl)) %>%
ggplot(mapping = aes(x = wt, y = mpg, colour = Cylinders)) +
geom_point(size = 3) +
labs(title = "Fuel Efficiency Declines as Weight Increases",
caption = "Urban Institute",
x = "Weight (Tons)",
y = "Miles Per Gallon") +
theme(legend.title = element_text(hjust = 0))
Bubble Scatter Plot
size = can be used as a mapping to plot a fourth dimension.
mtcars %>%
mutate(Cylinders = factor(cyl), `Automatic Transmission` = factor(am)) %>%
ggplot(mapping = aes(x = wt, y = mpg, color = `Automatic Transmission`, size = Cylinders)) +
geom_point() +
labs(title = "Fuel Efficiency Declines as Weight Increases",
caption = "Urban Institute",
x = "Weight (Tons)",
y = "Miles Per Gallon") +
theme(legend.title = element_text(hjust = 0))
Line Plots
ggplot(data = economics, mapping = aes(x = date, y = unemploy)) +
geom_line() +
scale_y_continuous(labels = scales::comma) +
labs(title = "Unemployment in the United States",
subtitle = "Number of Unemployed Americans in the U.S.",
caption = "Urban Institute",
x = "Year",
y = "Number Unemployed (1,000s)")
Lines Plots With Multiple Lines
library(tidyverse)
library(gapminder)
gapminder %>%
filter(country %in% c("Australia", "Canada", "New Zealand", "Japan")) %>%
ggplot(aes(year, gdpPercap, color = country)) +
geom_line() +
scale_y_continuous(expand = c(0, 0),
labels = scales::dollar) +
labs(title = "Per Capita GDP in Selected Countries",
subtitle = "From the gapminder data set",
caption = "Urban Institute",
x = "Year",
y = "Per capita GDP")
Plotting more than one variable can be useful for seeing the relationship of variables over time, but it takes a small amount of data munging.
This is because ggplot2 wants data in a “long” format instead of a “wide” format for line plots with multiple lines. gather() and spread() from the tidyr package make switching back-and-forth between “long” and “wide” painless. Essentially, variable titles go into “key” and variable values go into “value”. Then ggplot2, turns the different levels of the key variable (population, unemployment) into colors.
library(tidyverse)
as_tibble(EuStockMarkets) %>%
mutate(date = time(EuStockMarkets)) %>%
gather(key = "key", value = "value", -date) %>%
ggplot(aes(date, value, color = key)) +
geom_line() +
scale_y_continuous(expand = c(0, 0),
labels = scales::dollar,
limits = c(0, 10000)) +
labs(title = "Major European Stock Indices",
subtitle = "Based on daily closing prices",
caption = "Urban Institute",
x = "Date",
y = "Value")
Slope plots
# https://www.bls.gov/lau/
library(ggrepel)
unemployment <- tibble(
time = c("October 2009", "October 2009", "October 2009", "August 2017", "August 2017", "August 2017"),
rate = c(7.4, 7.1, 10.0, 3.9, 3.8, 6.4),
state = c("Maryland", "Virginia", "Washington D.C.", "Maryland", "Virginia", "Washington D.C.")
)
label <- tibble(label = c("October 2009", "August 2017"))
october <- filter(unemployment, time == "October 2009")
august <- filter(unemployment, time == "August 2017")
unemployment %>%
mutate(time = factor(time, levels = c("October 2009", "August 2017"))) %>%
ggplot() +
geom_line(aes(time, rate, group = state, color = state)) +
geom_point(aes(x = time, y = rate, color = state)) +
labs(subtitle = "Unemployment Rate",
caption = "Source: BLS Local Area Unemployment Statistics") +
theme(axis.ticks.x = element_blank(),
axis.title.x = element_blank(),
axis.ticks.y = element_blank(),
axis.title.y = element_blank(),
axis.text.y = element_blank(),
panel.grid.major.y = element_blank(),
panel.grid.minor.y = element_blank(),
panel.grid.major.x = element_blank(),
axis.line = element_blank()) +
geom_text_repel(data = october, mapping = aes(x = time, y = rate, label = as.character(rate)), nudge_x = -0.06) +
geom_text_repel(data = august, mapping = aes(x = time, y = rate, label = as.character(rate)), nudge_x = 0.06)
Binning
Histograms and kernel density polygons show the distribution of a single numeric variable. Histograms display distributions with a bar as a geom. Kernel density polygons compute and draw smooth kernel density estimates.
Histogram
ggplot(data = diamonds, mapping = aes(x = depth)) +
geom_histogram(bins = 100) +
scale_y_continuous(expand = c(0, 0), labels = scales::comma) +
labs(title = "Distribution of Diamond Depths",
caption = "Urban Institute",
x = "Depth",
y = "Count")
diamonds %>%
filter(cut == "Fair" | cut == "Premium") %>%
ggplot(mapping = aes(x = depth, fill = cut)) +
geom_histogram(bins = 100) +
scale_y_continuous(expand = c(0, 0), labels = scales::comma) +
labs(title = "Distribution of Diamond Depths and Quality",
caption = "Urban Institute",
x = "Depth",
y = "Count")
Smoothed kernel density estimates
population %>%
filter(year == 2013 & population < 3000000) %>%
ggplot(mapping = aes(x = population)) +
geom_density() +
scale_y_continuous(expand = c(0, 0)) +
scale_x_continuous(labels = scales::comma) +
labs(title = "Populations of World Countries",
caption = "Urban Institute",
x = "Population",
y = "Count")
diamonds %>%
filter(cut == "Ideal" | cut == "Premium") %>%
ggplot(aes(x = price, fill = cut)) +
geom_density(alpha = 0.2) +
scale_x_continuous(labels = scales::comma) +
scale_y_continuous(labels = scales::dollar) +
labs(title = "Distribution of Diamond Prices and Diamond Cut",
caption = "Urban Institue",
x = "Price",
y = "Count")
Ridgeline plots
Ridgeline plots are partially overlapping smoothed kernel density plots faceted by a categorical variable that pack a lot of information into one elegant plot.
library(ggridges)
ggplot(diamonds, mapping = aes(x = price, y = cut)) +
geom_density_ridges(fill = "#1696d2") +
scale_x_continuous(labels = scales::dollar) +
labs(title = "Distribution of Diamond Prices by Diamond Cut",
caption = "Urban Institue",
x = "Price",
y = "Cut")
Categorical Variables
The following plots are effective at displaying how a continuous or discrete variable varies across different levels of a categorical variable.
Box Plot
ggplot(InsectSprays, aes(x = spray, y = count)) +
geom_boxplot() +
scale_y_continuous(expand = c(0, 0)) +
labs(title = "Number of Insects Killed by Insect Sprays",
caption = "Urban Institute",
x = "Type of Insect Spray",
y = "Number of Dead Insects")
Violin Plot
ggplot(InsectSprays, aes(x = spray, y = count, fill = spray)) +
geom_violin() +
scale_y_continuous(expand = c(0, 0)) +
labs(title = "Number of Insects Killed by Insect Sprays",
caption = "Urban Institute",
x = "Type of Insect Spray",
y = "Number of Dead Insects")
Binned dot plots
TODO(aaron):
Bean Plots
TODO(aaron): Add bean plots
Area Plot
Stacked Area
txhousing %>%
filter(city %in% c("Austin","Houston","Dallas","San Antonio","Fort Worth")) %>%
group_by(city, year) %>%
summarize(sales = sum(sales)) %>%
ggplot(aes(x = year, y = sales, fill = city)) +
geom_area(position = "stack") +
scale_y_continuous(expand = c(0, 0), labels = scales::comma) +
labs(title = "Home Sales in Texas Cities",
caption = "Urban Institute",
x = "Year",
y = "Home Sales")
Filled Area
txhousing %>%
filter(city == "Austin" | city == "Houston"| city == "Dallas"| city == "San Antonio" | city == "Fort Worth") %>%
group_by(city, year) %>%
summarize(listings = sum(listings)) %>%
mutate(listings = ifelse(is.na(listings), lag(listings), listings)) %>%
ggplot(aes(x = year, y = listings, fill = city)) +
geom_area(position = "fill") +
scale_y_continuous(expand = c(0, 0), labels = scales::comma) +
labs(title = "Home Listings in Texas Cities",
caption = "Urban Institute",
x = "Year",
y = "Proportion of Home Listings")
Waffle Chart / Square Pie Chart
The waffle package {CRAN and Github} creates suare pie charts. It can also be combined with glyphs for more elegant shapes than squares. This example uses data pulled from A Vision for an Equitable DC.
Waffle charts will require a little extra tinkering since they are called from library(waffle) instead of library(ggplot2). Most importantly, waffle charts require theme_urban(text = element_text(family = "Lato")) for the Lato font.
Single Waffle Chart
library(waffle)
parts <- c(`Virginia\nClinics` = (1000 - 208 - 105), `Maryland\nClinics` = 208, `D.C.\nClinics` = 105)
waffle(parts, rows = 25, size = 1, colors = c("#1696d2", "#fdbf11", "#000000"), legend_pos = "bottom") +
labs(title = "Free Clinics in the D.C.-Maryland-Virginia Area",
subtitle = "1 Square == 1 Clinic",
caption = "Urban Institute") +
theme(text = element_text(family = "Lato"))
Waffle Charts with Glyphs
Waffle charts can be enhanced by replacing squares qith glyphs. Two important arguments to know are glyph_size = and use_glyph =. Both are called in the waffle() function. Note: size = 1 is sensible and glyph_size = 12 is sensible.
Using glyphs requires downloading fontawesome. That can be done here. Then run library(extrafont), port(<font-location>), and loadfonts() once. After that, building waffle charts with glpyhs should be as easy as one function call.
#library(extrafont)
#font_import("H:/IT/urban_R_theme/docs")
#loadfonts()
parts <- c(`Virginia\nClinics` = (50 - 10 - 5), `Maryland\nClinics` = 10, `D.C.\nClinics` = 5)
waffle(parts, rows = 5, glyph_size = 12, colors = c("#1696d2", "#fdbf11", "#000000"), legend_pos = "bottom", use_glyph = "medkit") +
labs(title = "Free Clinics in the D.C.-Maryland-Virginia Area",
subtitle = "1 Square == 20 Clinics",
caption = "Urban Institute") +
theme(text = element_text(family = "Lato"))
Multiple Waffle Charts
library(waffle) allows multiple waffle charts to be ironed together using iron(). Ironing multiple charts together requires some trial-and-error to get the sizes and resolution to look good, but the results can be worth the work. Don’t forget theme(text = element_text(family = "Lato"))!
library(waffle)
white <- c(`With Degree` = 169300, `Without Degree` = 800)
black <- c(`With Degree` = 174900, `Without Degree` = 34700)
hispanic <- c(`With Degree` = 27700, `Without Degree` = 12400)
iron(
waffle(white / 83, rows = 40, size = 0.25, colors = c("#1696d2", "#fdbf11"), title = "White", keep = FALSE, pad = 10) +
theme(text = element_text(family = "Lato")),
waffle(black / 83, rows = 40, size = 0.25, colors = c("#1696d2", "#fdbf11"), title = "Black", keep = FALSE) +
theme(text = element_text(family = "Lato")),
waffle(hispanic / 83, rows = 40, size = 0.25, colors = c("#1696d2", "#fdbf11"), title = "Hispanic", keep = FALSE, pad = 59, xlab = "1 Square == 83 People") +
theme(text = element_text(family = "Lato"))
) 
Heat map
library(fivethirtyeight)
bad_drivers %>%
mutate(`Number of Drivers` = scale(num_drivers),
`Percent Speeding` = scale(perc_speeding),
`Percent Alcohol` = scale(perc_alcohol),
`Percent Not Distracted` = scale(perc_not_distracted),
`Percent No Previous` = scale(perc_no_previous),
state = factor(state, levels = rev(state))
) %>%
select(-insurance_premiums, -losses, -(num_drivers:losses)) %>%
gather(`Number of Drivers`:`Percent No Previous`, key = "variable", value = "SD's from Mean") %>%
ggplot(aes(variable, state)) +
geom_tile(aes(fill = `SD's from Mean`)) +
labs(title = "Drivers Involved in Fatal Collisions By Behavior",
subtitle = "As a share of scaled fatal collisions per billion miles, 2009",
caption = "Source: fivethirtyeight R package",
x = NULL,
y = NULL) +
scale_fill_gradientn() +
theme(legend.position = "right",
legend.direction = "vertical",
axis.text.x = element_text(angle = 45))
#https://learnr.wordpress.com/2010/01/26/ggplot2-quick-heatmap-plotting/Faceting and Small Multiples
facet_wrap()
R’s faceting system is a powerful way to make “small multiples”.
Some edits to the theme may be necessary depending upon how many rows and columns are in the plot.
# Facet Wrap
txhousing %>%
filter(!city %in% c("Brazoria County", "Brownsville", "San Angelo", "Denton County")) %>%
ggplot(aes(x = median)) +
geom_histogram() +
facet_wrap(~city) +
scale_x_continuous(labels = scales::dollar) +
scale_y_continuous(expand = c(0, 0)) +
labs(title = "Median Monthly Home Prices in Selected Texas Cities",
x = "Median Monthly Home Value",
y = "Count") +
theme(axis.text.x = element_text(angle = 90),
strip.text = element_text(face = "plain",
size = rel(0.5)))
facet_grid()
## Facet Grid
mtcars %>%
mutate(vs = factor(vs, levels = 0:1, labels = c("Normal Transmission", "V/S Transmission")),
am = factor(am, levels = 0:1, labels = c("Automatic Transmission", "Manual Transmission"))) %>%
ggplot(aes(wt, mpg)) +
geom_point(alpha = 0.5) +
scale_y_continuous(expand = c(0, 0), limits = c(0, 45)) +
labs(title = "Determinants of Fuel Efficiency",
subtitle = "1974 Motor Trends US magazine data",
caption = "Urban Institute") +
facet_grid(vs ~ am, margins = TRUE) +
theme(panel.border = element_rect(colour = "black", fill = NA, size = 0.3))
Smoothers
geom_smooth() fits and plots models to data with two or more dimensions.
ggplot(data = mpg) +
geom_smooth(mapping = aes(x = displ, y = hwy))
Understanding and manipulating defaults is more important for geom_smooth() than other geoms because it contains a number of assumptions. geom_smooth() automatically uses loess for datasets with fewer than 1,000 observations and a generalized additive model with formula = y ~ s(x, bs = "cs") for datasets with greater than 1,000 observations. Both default to a 95% confidence interval with the confidence interval displayed.
Models are chosen with method = and can be set to lm(), glm(), gam(), loess(), rlm(), and more. Formulas can be specified with formula = and y ~ x syntax. Plotting the standard error is toggled with se = TRUE and se = FALSE, and level is specificed with level =. As always, more information can be seen in RStudio with ?geom_smooth().
geom_point() adds a scatterplot to geom_smooth(). The order of the function calls is important. The function called second will be layed on top of the function called first.
ggplot(data = diamonds, mapping = aes(x = carat, y = price)) +
geom_point(alpha = 0.1) +
geom_smooth(color = "#ec008b") +
scale_y_continuous(expand = c(0, 0), labels = scales::dollar)
geom_smooth can be subset by categorical and factor variables. This requires subgroups to have a decent number of observations and and a fair amount of variability across the x-axis. Confidence intervals often widen at the ends so special care is needed for the chart to be meaningful and readable.
ggplot(data = mpg, mapping = aes(x = displ, y = hwy, color = factor(cyl))) +
geom_point(alpha = 0.2) +
geom_smooth() +
labs(title = "Engine Displacement and City MPG by Number of Cylinders",
subtitle = "Loess: MPG = Displacement",
caption = "Urban Institute",
x = "Engine Displacement",
y = "Highway MPG")
ggplot(data = mpg, mapping = aes(x = displ, y = hwy, color = drv)) +
geom_smooth(method = "lm") +
geom_point(alpha = 0.2) +
labs(title = "Engine Displacement and City MPG by Drive",
subtitle = "Linear Model: MPG = Displacement",
caption = "Urban Institute",
x = "Engine Displacement",
y = "Highway MPG")
Highlighting
points polygons lines
Text
library(ggplot2)
geom_text() and geom_label() annotate points on scatterplots. Both functions need x values, y values, and a vector of labels. It’s rarely useful to label every point. Use filter() to create a second data set that is subsetted and pass it into the labelling function.
geom_text() labels points without backgrounds. geom_label() labels points with backgrounds.
geom_text()
labels <- mtcars %>%
rownames_to_column("model") %>%
filter(model %in% c("Toyota Corolla", "Merc 240D", "Datsun 710"))
mtcars %>%
ggplot(mapping = aes(x = wt, y = mpg)) +
geom_point() +
geom_text(data = labels, mapping = aes(x = wt, y = mpg, label = model), nudge_x = 0.38) +
labs(title = "Fuel Efficiency Declines as Weight Increases",
caption = "Urban Institute",
x = "Weight (Tons)",
y = "Miles Per Gallon")
geom_label()
labels <- mtcars %>%
rownames_to_column("model") %>%
filter(model %in% c("Toyota Corolla", "Merc 240D", "Datsun 710"))
mtcars %>%
ggplot(mapping = aes(x = wt, y = mpg)) +
geom_point() +
geom_label(data = labels, mapping = aes(x = wt, y = mpg, label = model), nudge_x = 0.35) +
labs(title = "Fuel Efficiency Declines as Weight Increases",
caption = "Urban Institute",
x = "Weight (Tons)",
y = "Miles Per Gallon")
library(ggrepel)
Text too often overlaps when using geom_text() or geom_label(). library(ggrepel) is a library(ggplot2) add-on that automatically positions text so it doesn’t overlap with geoms or other text.
To add this functionalit simple load library(ggrepel) and then use geom_text_repel() and geom_text_label() with the same syntax as geom_text() and geom_label().
geom_text_repel()
library(ggrepel)
labels <- mtcars %>%
rownames_to_column("model") %>%
filter(model %in% c("Toyota Corolla", "Merc 240D", "Datsun 710"))
mtcars %>%
ggplot(mapping = aes(x = wt, y = mpg)) +
geom_point() +
geom_text_repel(data = labels, mapping = aes(x = wt, y = mpg, label = model), nudge_x = 0.38) +
labs(title = "Fuel Efficiency Declines as Weight Increases",
caption = "Urban Institute",
x = "Weight (Tons)",
y = "Miles Per Gallon")
geom_label_repel()
labels <- mtcars %>%
rownames_to_column("model") %>%
filter(model %in% c("Toyota Corolla", "Merc 240D", "Datsun 710"))
mtcars %>%
ggplot(mapping = aes(x = wt, y = mpg)) +
geom_point() +
geom_label_repel(data = labels, mapping = aes(x = wt, y = mpg, label = model), nudge_x = 0.35) +
labs(title = "Fuel Efficiency Declines as Weight Increases",
caption = "Urban Institute",
x = "Weight (Tons)",
y = "Miles Per Gallon")
Layered Geoms
Geoms can be layered in ggplot2. This is useful for design and analysis.
It is often useful to add points to line plots with a small number of values across the x-axis. This example from R for Data Science shows how changing the line to grey can be appealing.
Design
Before
table1 %>%
ggplot(aes(x = year, y = cases)) +
geom_line(aes(color = country)) +
geom_point(aes(color = country)) +
scale_y_continuous(labels = scales::comma) +
scale_x_continuous(breaks = c(1999, 2000)) +
labs(title = "Changes in Tuberculosis Cases in Three Countries",
caption = "Source: World Health Organization Global Tuberculosis Report")
After
table1 %>%
ggplot(aes(year, cases)) +
geom_line(aes(group = country), color = "grey50") +
geom_point(aes(color = country)) +
scale_y_continuous(labels = scales::comma) +
scale_x_continuous(breaks = c(1999, 2000)) +
labs(title = "Changes in Tuberculosis Cases in Three Countries",
caption = "Source: World Health Organization Global Tuberculosis Report")
Layering geoms is also useful for adding trend lines and centroids to scatter plots.
# Simple line
# Regression model
# CentroidsCentroids
mpg_summary <- mpg %>%
group_by(cyl) %>%
summarize(displ = mean(displ), cty = mean(cty))
mpg %>%
ggplot() +
geom_point(aes(x = displ, y = cty, color = factor(cyl)), alpha = 0.5) +
geom_point(data = mpg_summary, aes(x = displ, y = cty), size = 5, color = "red") +
geom_text(data = mpg_summary, aes(x = displ, y = cty, label = cyl)) +
labs(title = "City MPG, Engine Displacement, and Number of Cylinders",
subtitle = "Subgroup Means in Red",
caption = "Source: EPA Fuel Economy Dataset",
x = "Displacement",
y = "City MPG")
Saving plots
Appendix
System Info and Package Versioning
sessionInfo()## R version 3.4.3 (2017-11-30)
## Platform: x86_64-apple-darwin15.6.0 (64-bit)
## Running under: macOS High Sierra 10.13.3
##
## Matrix products: default
## BLAS: /Library/Frameworks/R.framework/Versions/3.4/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/3.4/Resources/lib/libRlapack.dylib
##
## locale:
## [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] fivethirtyeight_0.3.0 waffle_0.7.0 ggridges_0.4.1
## [4] gapminder_0.3.0 hexbin_1.27.2 bindrcpp_0.2
## [7] ggrepel_0.7.0 extrafont_0.17 forcats_0.2.0
## [10] stringr_1.2.0 dplyr_0.7.4 purrr_0.2.4
## [13] readr_1.1.1 tidyr_0.8.0 tibble_1.4.2
## [16] ggplot2_2.2.1 tidyverse_1.2.1 knitr_1.19
##
## loaded via a namespace (and not attached):
## [1] Rcpp_0.12.15 lubridate_1.7.1 lattice_0.20-35
## [4] png_0.1-7 assertthat_0.2.0 rprojroot_1.3-2
## [7] digest_0.6.15 psych_1.7.8 R6_2.2.2
## [10] cellranger_1.1.0 plyr_1.8.4 backports_1.1.2
## [13] evaluate_0.10.1 httr_1.3.1 pillar_1.1.0
## [16] rlang_0.1.6 lazyeval_0.2.1 readxl_1.0.0
## [19] rstudioapi_0.7 extrafontdb_1.0 Matrix_1.2-12
## [22] rmarkdown_1.8 labeling_0.3 foreign_0.8-69
## [25] munsell_0.4.3 broom_0.4.3 compiler_3.4.3
## [28] modelr_0.1.1 pkgconfig_2.0.1 mnormt_1.5-5
## [31] mgcv_1.8-23 htmltools_0.3.6 tidyselect_0.2.3
## [34] gridExtra_2.3 crayon_1.3.4 grid_3.4.3
## [37] nlme_3.1-131 jsonlite_1.5 Rttf2pt1_1.3.5
## [40] gtable_0.2.0 magrittr_1.5 scales_0.5.0
## [43] cli_1.0.0 stringi_1.1.6 reshape2_1.4.3
## [46] xml2_1.2.0 RColorBrewer_1.1-2 tools_3.4.3
## [49] glue_1.2.0 hms_0.4.1 parallel_3.4.3
## [52] yaml_2.1.16 colorspace_1.3-2 rvest_0.3.2
## [55] bindr_0.1 haven_1.1.1